John Cox

Blogs · @wyome

Tagged claude · everything

Claude Code Docs

Best practices for Claude Code - Claude Code Docs

This is Anthropic's best-practices guide for Claude Code, and its central argument is that nearly every recommendation follows from one constraint: the context window fills quickly and performance degrades as it does. From that premise it lays out how to work with an agent that acts autonomously, including giving Claude a verification check so it can close its own loop, exploring and planning before coding, configuring CLAUDE.md, permissions, hooks, and skills, and managing context through /clear, compaction, and subagents. It is worth keeping as a practical reference for anyone moving from prompting a chatbot to delegating multi-step coding work to an agent.

Permalink productivityprogrammingaitips

optimAIzr

optimAIzr - spend fewer tokens on the same work

This is the landing page for optimaizr, a free, open-source MIT-licensed CLI that reads token counts from the usage files Claude Code, Codex and opencode already write, then prices wasted tokens in dollars and applies accepted fixes such as compacting conversations earlier. Its headline evidence is the author's own profile: 55% of a $1,020 API-equivalent bill went to re-reading conversation context, and compacting at 200K reportedly cut cost per prompt by 26%, at the price of about seven minutes a day waiting for summaries. The page distinguishes "clear waste" from "likely" and "up to" savings and states that usage data stays on the machine, which is what sets it apart from other cost-tracking tools.

Permalink aiclaudeopensource

Claude

Build live dashboards and animate explainers with Claude | Claude by Anthropic

Anthropic's announcement introduces Claude Dashboards (live, query-backed dashboards built from company data platforms and CRM tools, in beta on paid plans) and Claude Motion (code-based explainer animations that can be edited and exported as MP4, in beta on Team and Enterprise), while graduating Docs, Slides, and Design out of beta for all plans. It also covers the retirement of the standalone claude.ai/design site, with a December 14 cutoff and a migration path for design systems and chats. What sets it apart from other launch posts is the emphasis on transparency and control: each chart exposes its underlying query, animations are editable code rather than generated video, and admins decide which features are enabled.

Permalink productivityanimationclaudedata

Claude

Customize Claude Code with mods in TypeScript | Claude by Anthropic

Anthropic has released mods for Claude Code, small TypeScript functions that let users customize how the AI coding assistant works by hooking into events like tool calls, permission requests, and UI rendering—users can rewrite prompts, block tool calls, redact secrets, or modify the interface without waiting for official features. Mods ship as plugins and are stackable (multiple mods can hook the same event), run with full machine access like any installed code, and teams can use them for governance features like production safeguards or audit logging, with enterprise controls available through a built-in security mod that restricts risky operations.

Permalink developmenttoolsclaudeplugins

www.reddit.com

www.reddit.com

www.reddit.com

Claude

What a task costs on Opus 5.5 | Claude by Anthropic

Claude Code tasks cost money based on four factors: the number of turns (which resend the conversation), cache hit rates (dramatically cheaper than fresh reads), output tokens including thinking (the most expensive element), and model choice. Opus 5.5 is 20% cheaper on input and output tokens and 60% cheaper on cached reads than Opus 5, shifting the economics significantly toward long, cache-heavy sessions. The post provides interactive calculators and concrete guidance: use Opus 5.5 as your daily driver for supervised work, raise effort level before switching models, put lookup tasks on smaller models like Sonnet, keep sessions moving to maintain cache warmth, and crucially, measure your own real tasks to understand how these tradeoffs apply to your specific codebase and workflow.

Permalink developmentclaudepricing

GitHub

GitHub - onmyway133/claude-code-tips-tricks: My favorite Claude Code tips & tricks · GitHub

A collection of tips for using Claude Code effectively, organized into eight categories covering sessions, command mode, agents, skills, MCP servers, prompting, hooks, and workflow management. Topics range from practical efficiency gains like naming sessions after git branches and using subagents for exploration to architectural choices like running parallel worktrees for isolation and using hooks as deterministic gates for unattended runs. The guide emphasizes that context is the real constraint in long sessions, so managing what stays in the conversation window through deliberate pruning, subagent delegation, and rewinding failed attempts matters more than piling on additional rules.

Permalink productivitydevelopmenttipsclaude

cuong.io

Claude Code Best Practices: Memory Management | Code Centre

Claude Code's memory management system emphasizes keeping instructions lean by storing only essential, session-agnostic guidelines in CLAUDE.md files while delegating project-specific details to a docs/ folder referenced as needed. Rather than using memory banks or frequent compacting (which wastes tokens and fidelity), users should structure workflows around architecture documents, implementation plans with checkboxes, and selective /compact usage only when context window pressure demands it while preserving session continuity.

Permalink productivitysoftwaredevelopmentadvice

Note

How-To Geek

I built a dynamic GitHub app store with Claude, now I don't have to rely on Google for open-source app discovery

The author built a custom Claude-powered app store interface that sits atop GitHub to address the platform's poor user experience for discovering and installing open-source software. Using Claude's web search and API access, the system curates 10-15 relevant applications from multiple sources, presents them in a clean gallery with screenshots and metadata, performs basic security checks on the source code, verifies system dependencies, and can automate installation—translating developer jargon into plain language for non-technical users. The article includes a detailed prompt to create this as a reusable Claude skill, demonstrating how LLMs can make technical tools accessible to anyone regardless of coding knowledge.

Permalink developmentautomationclaudeopen source

help.dropbox.com

How to connect Dropbox to Claude

This is Dropbox's own support/setup guide for its Claude integration, covering both the Dropbox connector (for claude.ai and Claude desktop) and the Dropbox plugin (for Claude Cowork and Code). It walks through connection steps, the specific permission scopes requested (account info, file/folder access, sharing, file requests), and admin controls for allowing or blocking the integration org-wide. Worth keeping as a reference for setup/troubleshooting and for understanding exactly what access Claude gains, including limits like text-only file saving and existing-permissions enforcement.

Permalink softwareclaude